Gaze-driven object tracking based on optical flow estimation

Abstract

To efficiently deploy eye tracking within gaze-dependent image analysis tasks, we present an optical flow-aided extension of the gaze-driven object tracking technique (GDOT). GDOT assumes that objects in a 3-dimensional space are fixation targets and with high probability computes the fixation directions towards the target observed by the user. We research whether this technique proves its efficiency for video footage in 2-dimensional space in which the targets are tracked by optical flow tracking technique with inaccuracies characteristic for this method. In the conducted perceptual experiments, we assess efficiency of the gazedriven object identification by comparing results with the reference data where attended objects are known. The GDOT extension reveals higher errors in comparison to 3D graphics tasks but still outperforms typical fixation techniques.